Appearance
Are you an LLM? You can read better optimized documentation at /mcp/market-analysis.md for this page in Markdown format
Market History & What-if Analysis
When a trade leaves you asking “what happened?” or “what if my target had been different?”, start with its recorded history. Then inspect the market around it and, where the evidence supports it, compare a hypothetical change.
Experimental — availability and evidence limits apply
This version enables the market-analysis extension by default; the server operator can explicitly disable it. This does not grant MCP account access or configure the market-data provider. Missing provider configuration returns an explanation, not invented candles. The interactive viewer requires a compatible MCP Apps client; text and tables remain available without it. This documentation does not certify a production rollout or compatibility with every Claude or ChatGPT client.
Three kinds of evidence
| Evidence | What it answers | What it cannot prove |
|---|---|---|
| Recorded trade and trace | What TTMT recorded: submitted orders, fills, target changes, management decisions and timing, where retained | Missing events or a different hypothetical outcome |
| External LSE candles | How the external market feed moved during a stated UTC window | Your broker’s exact prices, spread or whether your order filled |
| Paired replay | How a declared model behaves with the historical baseline and your alternatives on the same eligible trades and frozen candles | What your broker would certainly have executed, future profitability or net returns after all costs |
London Strategic Edge (LSE) provides the candle history. Every analysis names its source, symbol mapping, time window, resolution, coverage and data fingerprint. If the feed does not establish whether prices are bid, ask or midpoint, the analysis says the price basis is unknown. Missing candles are not invented or interpolated.
Trace is useful for more than a narrative. Where recorded evidence is sufficient, it can establish the original order plan and historical settings used by the replay. The assistant can inspect recorded trace events separately from the replay’s simulated events. Missing or ambiguous evidence is a limitation, not permission to substitute today’s settings.
Scenario: TP1 at 150 instead of 100 pips
Example question
“On my demo account, for Channel X’s trades last week, compare replacing TP1 with 150 pips against the historical baseline. Use a 24-hour replay horizon. Show the matched sample, exclusions, baseline mismatch and uncertainty.”
Name an account, channel and date window. The horizon is how long each hypothetical trade is followed from its start; it is not cut short at the real trade’s closing time. Your assistant should resolve unclear account names, pip definitions or missing scope before calculating.
“Make TP1 150” has two different meanings:
- Replace TP1 directly: forcibly replace the historical first target in the hypothetical plan.
- Change a TTMT setting: change a named account or profile setting while respecting its historical prerequisites and precedence. A signal-provided target or profile override may mean your change has no effect.
Neither action changes your saved settings. The result reports which trades the change affected and why others were unchanged or excluded. Do not silently assume every historical TP1 was 100 pips: check the baseline evidence first. If you explicitly want a modeled 100-versus-150 comparison, ask for that baseline as a separate hypothetical scenario.
Read the comparison
Keep these views separate:
- Actual full cohort: recorded outcomes for all trades in your selected scope.
- Actual matched cohort: recorded outcomes for the subset eligible for every requested scenario.
- Replayed baseline: the baseline evaluated by the model on that common subset.
- Replayed alternatives: your changes evaluated on exactly the same subset and market data.
Compare an alternative with the replayed baseline, not directly with the full actual results. Baseline mismatch shows where the model differs from recorded history. A large mismatch weakens the usefulness of its hypothetical comparison.
The baseline assessment distinguishes different completion states, a numeric net-versus-gross difference, no detected numeric difference, and unassessed. Missing risk data or an exit beyond the replay window is not agreement. Even equal R does not establish matching fills, timing or costs; none of these states diagnoses a broker error.
The viewer separates trades affected by a policy change from outcomes classified as improved, worsened, unchanged or unresolved. Expand Inspect trade outcomes beneath an alternative to open a trade with that scenario selected. These classifications describe the primary modeled path; sensitivity can change their direction.
Every metric carries its sample or denominator. Open-at-horizon trades are not silently counted as wins or losses. Gross modeled R is not actual net account profit. Monetary results stay within one account and currency; a cumulative modeled-R chart is not account equity or drawdown.
If changing the assumed order of price moves within a candle or management latency changes the direction of the result, treat the answer as inconclusive. Where available, a bounded finer-candle pass improves event ordering. It only replaces complete intervals that agree with the coarser candles; missing, partial or conflicting fine history remains disclosed. It is not an exhaustive tick reconstruction.
Follow up without moving the comparison
Ask: “Now try 175 pips on the same trades and frozen data.” The assistant reuses the analysis context instead of fetching a different market window. A new immutable result records the alternative. If you narrow the trade set, the result discloses the removed trades.
Tested alternatives in this analysis chain shows up to 12 prior runs with their baseline, alternatives, context identifiers and matched/requested counts. Older omitted runs are counted explicitly. The same journal is included in text results and export context. It covers only that follow-up ancestry, not other branches or separately started analyses; different contexts are not paired comparisons. Trying several alternatives and presenting only the winner is not evidence of a validated strategy.
Analyses last at most 30 minutes from the original context’s creation. Reading them or asking follow-ups does not extend that deadline. They can become unavailable earlier after server restart, eviction or loss of access to their underlying trades. An expired result cannot silently refresh under the same identifier; start a new analysis and treat it as a new comparison.
Inspect charts, events and data
Ask “Show this analysis” to open the optional viewer. In compatible clients you can inspect a comparison, select a trade, view genuine base or finer candles, and read recorded trace and simulated-event details separately. The viewer reads the same stored result as the assistant’s text answer; it does not run a second calculation in your browser.
Missing intervals within the displayed candles appear as blank chart spaces with an expandable UTC interval list. Long gaps are compressed on the time axis; a blank space is not evidence of a market closure.
If your client cannot render the viewer, ask for its summary, a trade detail, candle rows or CSV through show_analysis. Candle, event and export views are paginated. A page is not the entire dataset: retain its provenance and follow the returned cursor when more rows remain. Exports do not make the private analysis publicly accessible.
The viewer exports CSV plus a companion context JSON file. Keep both: the context records the analysis scope, scenarios, historical units, provenance, assumptions and metric denominators needed to interpret the rows. Large exports use bounded batches; choose Export next batch until no rows remain, and retain every earlier batch. Each context file states its row range and whether it contains the whole export. Completing an export does not establish complete market coverage. If your host cannot download files, the viewer provides both texts to copy. An analysis expiring does not delete copies you already downloaded.
Anthropic’s MCP Apps documentation describes testing an App’s inline display in Claude. OpenAI’s connection and testing guide covers remote MCP connections and optional UI validation in ChatGPT. These are host capabilities, not evidence that this TTMT extension has completed its release checks.
Limits and unsupported cases
An initial replay supports up to 100 trades, six symbols, three alternatives and a 30-day selection window, with a maximum 72-hour horizon per trade. Provider work is bounded by request, time, row and byte budgets. A narrower window can help a budget error; it cannot fix missing evidence.
The model supports evidenced original market/limit plans, fixed targets and stops, supported TP-triggered breakeven and strategy-aware target redistribution. It is not a full broker emulator. Trailing stops, open-target runners, trade-level money limits, portfolio close-all and material recorded interventions can make a trade unsupported. Missing historical pip units, settings, original orders or causal timestamps also prevent a trustworthy replay. The result lists reasons rather than quietly dropping those trades.
You can still ask for market history without replay when candle access is available but replay evidence is insufficient.
New history and replay calls consume your normal MCP daily allowance, including follow-up calculations. Inspecting an already-created analysis with show_analysis does not consume that daily allowance; authentication, access checks and per-minute limits still apply.
If something goes wrong
| State | Next step |
|---|---|
| Tools are not listed | The extension is disabled on that server. Use recorded-trade tools; do not ask the assistant to invent a replay. |
| Provider unavailable or rate-limited | Retry later. An error is not a zero-return result. |
| Symbol or historical evidence unsupported | Read the reason. Inspect the recorded trace or candles separately; do not guess missing inputs. |
| Partial coverage or unresolved outcomes | Review the missing intervals, exclusions and sensitivity before drawing a conclusion. |
| Analysis unavailable | It may have expired, been evicted or lost access. Start a new analysis if you still need it. |
| No interactive viewer | Request the text summary, tables or bounded export instead. |